Skip to content

Fixes for GH200 hardware compilation (+ misc) #183

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
FlorianDeconinck merged 4 commits into from
Jul 21, 2025
Merged

Fixes for GH200 hardware compilation (+ misc) #183

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
3025cd5
Align logging default to usage in Pace
FlorianDeconinck Jul 18, 2025
c2454e0
Fix compile flags for GH boxes
FlorianDeconinck Jul 18, 2025
f25debf
Merge branch 'develop' into gix/GH200
FlorianDeconinck Jul 18, 2025
22573f3
Update ndsl/dsl/dace/dace_config.py
FlorianDeconinck Jul 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 14 additions & 1 deletion ndsl/dsl/dace/dace_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -204,12 +204,25 @@ def __init__(
"openmp_sections",
value=0,
)
# Resolve "march/mtune" option
# - turn numeric-centric SSE by default
# - Neoverse-V2 Grace CPU will fail - use alternative mcpu=native.
# Detecting neoverse-v1/2 requires an external package, we swap it
# for a read on GH200 nodes themselves
march_option = "-Xcompiler -march=native"
if (
cp is not None
and cp.cuda.runtime.getDeviceProperties(0)["name"]
== b"NVIDIA GH200 480GB"
):
march_option = "-Xcompiler -mcpu=native"

# Removed --fast-math
dace.config.Config.set(
"compiler",
"cuda",
"args",
value="-std=c++14 -Xcompiler -fPIC -O3 -Xcompiler -march=native",
value=f"-std=c++14 -Xcompiler -fPIC -O3 {march_option}",
)

cuda_sm = 60
Expand Down
10 changes: 5 additions & 5 deletions ndsl/logging.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
from mpi4py import MPI


LOGLEVEL = os.environ.get("PACE_LOGLEVEL", "INFO").upper()
LOGLEVEL = os.environ.get("PACE_LOGLEVEL", "INFO").lower()

# Python log levels are hierarchical, therefore setting INFO
# means DEBUG and everything lower will be logged.
Expand All @@ -23,10 +23,10 @@

def _ndsl_logger() -> logging.Logger:
name_log = logging.getLogger(__name__)
name_log.setLevel(LOGLEVEL)
name_log.setLevel(AVAILABLE_LOG_LEVELS[LOGLEVEL])

handler = logging.StreamHandler(sys.stdout)
handler.setLevel(LOGLEVEL)
handler.setLevel(AVAILABLE_LOG_LEVELS[LOGLEVEL])
formatter = logging.Formatter(
fmt=(
f"%(asctime)s|%(levelname)s|rank {MPI.COMM_WORLD.Get_rank()}|"
Expand All @@ -41,13 +41,13 @@ def _ndsl_logger() -> logging.Logger:

def _ndsl_logger_on_rank_0() -> logging.Logger:
name_log = logging.getLogger(f"{__name__}_on_rank_0")
name_log.setLevel(LOGLEVEL)
name_log.setLevel(AVAILABLE_LOG_LEVELS[LOGLEVEL])

rank = MPI.COMM_WORLD.Get_rank()

if rank == 0:
handler = logging.StreamHandler(sys.stdout)
handler.setLevel(LOGLEVEL)
handler.setLevel(AVAILABLE_LOG_LEVELS[LOGLEVEL])
formatter = logging.Formatter(
fmt=(
f"%(asctime)s|%(levelname)s|rank {MPI.COMM_WORLD.Get_rank()}|"
Expand Down